A Battery State-of-Health Estimation Method for Real-World Electric Vehicles Based on Physics-Informed Neural Networks

Accurate estimation of the battery state of health (SOH) in real-world electric vehicles (EVs), utilizing sensor data, is crucial for ensuring both reliability and efficiency. This article proposes a battery SOH estimation scheme designed to operate under conditions with limited data. Initially, an improved battery SOH calculation method based on the incremental capacity (IC) curves is proposed to adapt to the multistage constant-current charging scenarios. Subsequently, the latent Dirichlet allocation (LDA) model is employed to cluster the driving topics and analyze their impact on the battery SOH. Finally, driving behaviors are incorporated as health features into a variance uncertainty-weighted physics-informed neural network (PINN) for the SOH estimation. The results show that the proposed model outperforms existing approaches, achieving a mean absolute percentage error (MAPE) and a root mean square percentage error (RMSPE) of 2.6862% and 2.9630%, respectively, at a relatively low computational cost. In addition, the impact of the physical constraints on the model is analyzed using Shapley values.